If all of the Q&A platforms die off, how are LLM training datasets going to get new information?
You just take arbitrary data and ask the LLM to put it in Q&A format and generate the synthetic training data. Unless you are suggesting Quora is the source of new information, which I don't agree with.
Quora does not care about the user experience. Their obsession with pay-walling killed the site for me across a decade. They literally could not get me to sign up and boy did they try (I really needed an answer once too!). My soul really remembers hostile sites.
In my experience, they do seem to be very good at synthesizing answers from docs. However I don't know if that will work for edge cases which is one of the things SO is good at.
Why do people keep repeating this falsehood? Is it wishful thinking, or a genuine technical misunderstanding, or intentional disinformation?
LLMs absolutely can create novel syntheses. It’s very easy to test this yourself. From creating sentences that do not appear in Google to creating unique story outlines, it’s super easy to prove this wrong.
I think it's a matter of perception. There's regurgitation. There's recombination. There's advanced recombination through layers of prestidigitation. And then there's actual human creativity, which you might deny is special, leaving us at an impasse because we can't provide you with a tool to measure it with. It just comes down to a philosophical face off, high noon with hand-waving instead of six-guns.
But anyway the point is that LLMs produce a lot of novel stuff that we feel already tired of because it seems like we've seen it before.
> It is frankly absurd that they should be expected to
> These LLMs could not exist without them, but now they're expected to compete?
Yea, those damn tractor makers - they ate the food that the hand farmers used to make! How are hand farmers expected to compete with tractors now, when it's so much more efficient and can do 100x the work!?
These LLMs could not exist without them, but now they're expected to compete?
If all of the Q&A platforms die off, how are LLM training datasets going to get new information?
This whole AI boom is typical corporate shortsightedness imo. Kill the future in order to have a great next quarter
I hope I'm wrong. If I am right, then I hope we figure this out before AI has bulldozed everything into dust